drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <stdio.h>
  37 #include <errno.h>
  38 #include <unistd.h>
  39
  40 #include <rte_ethdev.h>
  41 #include <rte_ethdev_pci.h>
  42 #include <rte_memcpy.h>
  43 #include <rte_string_fns.h>
  44 #include <rte_memzone.h>
  45 #include <rte_malloc.h>
  46 #include <rte_atomic.h>
  47 #include <rte_branch_prediction.h>
  48 #include <rte_pci.h>
  49 #include <rte_ether.h>
  50 #include <rte_common.h>
  51 #include <rte_errno.h>
  52 #include <rte_cpuflags.h>
  53
  54 #include <rte_memory.h>
  55 #include <rte_eal.h>
  56 #include <rte_dev.h>
  57
  58 #include "virtio_ethdev.h"
  59 #include "virtio_pci.h"
  60 #include "virtio_logs.h"
  61 #include "virtqueue.h"
  62 #include "virtio_rxtx.h"
  63
  64 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  65 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  66 static int  virtio_dev_start(struct rte_eth_dev *dev);
  67 static void virtio_dev_stop(struct rte_eth_dev *dev);
  68 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  69 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  70 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  71 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  72 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  73                                 struct rte_eth_dev_info *dev_info);
  74 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  75         int wait_to_complete);
  76 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  77
  78 static void virtio_set_hwaddr(struct virtio_hw *hw);
  79 static void virtio_get_hwaddr(struct virtio_hw *hw);
  80
  81 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  82                                  struct rte_eth_stats *stats);
  83 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  84                                  struct rte_eth_xstat *xstats, unsigned n);
  85 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  86                                        struct rte_eth_xstat_name *xstats_names,
  87                                        unsigned limit);
  88 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  89 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  90 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  91                                 uint16_t vlan_id, int on);
  92 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  93                                 struct ether_addr *mac_addr,
  94                                 uint32_t index, uint32_t vmdq);
  95 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  96 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  97                                 struct ether_addr *mac_addr);
  98
  99 static int virtio_dev_queue_stats_mapping_set(
 100         struct rte_eth_dev *eth_dev,
 101         uint16_t queue_id,
 102         uint8_t stat_idx,
 103         uint8_t is_rx);
 104
 105 /*
 106  * The set of PCI devices this driver supports
 107  */
 108 static const struct rte_pci_id pci_id_virtio_map[] = {
 109         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
 110         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
 111         { .vendor_id = 0, /* sentinel */ },
 112 };
 113
 114 struct rte_virtio_xstats_name_off {
 115         char name[RTE_ETH_XSTATS_NAME_SIZE];
 116         unsigned offset;
 117 };
 118
 119 /* [rt]x_qX_ is prepended to the name string here */
 120 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 121         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 122         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 123         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 124         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 125         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 126         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 127         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 128         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 129         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 130         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 131         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 132         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 133         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 134 };
 135
 136 /* [rt]x_qX_ is prepended to the name string here */
 137 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 138         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 139         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 140         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 141         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 142         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 143         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 144         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 145         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 146         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 147         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 148         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 149         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 150         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 151 };
 152
 153 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 154                             sizeof(rte_virtio_rxq_stat_strings[0]))
 155 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 156                             sizeof(rte_virtio_txq_stat_strings[0]))
 157
 158 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 159
 160 static int
 161 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 162                 int *dlen, int pkt_num)
 163 {
 164         uint32_t head, i;
 165         int k, sum = 0;
 166         virtio_net_ctrl_ack status = ~0;
 167         struct virtio_pmd_ctrl *result;
 168         struct virtqueue *vq;
 169
 170         ctrl->status = status;
 171
 172         if (!cvq || !cvq->vq) {
 173                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 174                 return -1;
 175         }
 176         vq = cvq->vq;
 177         head = vq->vq_desc_head_idx;
 178
 179         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 180                 "vq->hw->cvq = %p vq = %p",
 181                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 182
 183         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 184                 return -1;
 185
 186         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 187                 sizeof(struct virtio_pmd_ctrl));
 188
 189         /*
 190          * Format is enforced in qemu code:
 191          * One TX packet for header;
 192          * At least one TX packet per argument;
 193          * One RX packet for ACK.
 194          */
 195         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 196         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 197         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 198         vq->vq_free_cnt--;
 199         i = vq->vq_ring.desc[head].next;
 200
 201         for (k = 0; k < pkt_num; k++) {
 202                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 203                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 204                         + sizeof(struct virtio_net_ctrl_hdr)
 205                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 206                 vq->vq_ring.desc[i].len = dlen[k];
 207                 sum += dlen[k];
 208                 vq->vq_free_cnt--;
 209                 i = vq->vq_ring.desc[i].next;
 210         }
 211
 212         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 213         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 214                         + sizeof(struct virtio_net_ctrl_hdr);
 215         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 216         vq->vq_free_cnt--;
 217
 218         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 219
 220         vq_update_avail_ring(vq, head);
 221         vq_update_avail_idx(vq);
 222
 223         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 224
 225         virtqueue_notify(vq);
 226
 227         rte_rmb();
 228         while (VIRTQUEUE_NUSED(vq) == 0) {
 229                 rte_rmb();
 230                 usleep(100);
 231         }
 232
 233         while (VIRTQUEUE_NUSED(vq)) {
 234                 uint32_t idx, desc_idx, used_idx;
 235                 struct vring_used_elem *uep;
 236
 237                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 238                                 & (vq->vq_nentries - 1));
 239                 uep = &vq->vq_ring.used->ring[used_idx];
 240                 idx = (uint32_t) uep->id;
 241                 desc_idx = idx;
 242
 243                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 244                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 245                         vq->vq_free_cnt++;
 246                 }
 247
 248                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 249                 vq->vq_desc_head_idx = idx;
 250
 251                 vq->vq_used_cons_idx++;
 252                 vq->vq_free_cnt++;
 253         }
 254
 255         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 256                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 257
 258         result = cvq->virtio_net_hdr_mz->addr;
 259
 260         return result->status;
 261 }
 262
 263 static int
 264 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 265 {
 266         struct virtio_hw *hw = dev->data->dev_private;
 267         struct virtio_pmd_ctrl ctrl;
 268         int dlen[1];
 269         int ret;
 270
 271         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 272         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 273         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 274
 275         dlen[0] = sizeof(uint16_t);
 276
 277         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 278         if (ret) {
 279                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 280                           "failed, this is too late now...");
 281                 return -EINVAL;
 282         }
 283
 284         return 0;
 285 }
 286
 287 static void
 288 virtio_dev_queue_release(void *queue __rte_unused)
 289 {
 290         /* do nothing */
 291 }
 292
 293 static int
 294 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
 295 {
 296         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
 297                 return VTNET_CQ;
 298         else if (vtpci_queue_idx % 2 == 0)
 299                 return VTNET_RQ;
 300         else
 301                 return VTNET_TQ;
 302 }
 303
 304 static uint16_t
 305 virtio_get_nr_vq(struct virtio_hw *hw)
 306 {
 307         uint16_t nr_vq = hw->max_queue_pairs * 2;
 308
 309         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 310                 nr_vq += 1;
 311
 312         return nr_vq;
 313 }
 314
 315 static void
 316 virtio_init_vring(struct virtqueue *vq)
 317 {
 318         int size = vq->vq_nentries;
 319         struct vring *vr = &vq->vq_ring;
 320         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 321
 322         PMD_INIT_FUNC_TRACE();
 323
 324         /*
 325          * Reinitialise since virtio port might have been stopped and restarted
 326          */
 327         memset(ring_mem, 0, vq->vq_ring_size);
 328         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
 329         vq->vq_used_cons_idx = 0;
 330         vq->vq_desc_head_idx = 0;
 331         vq->vq_avail_idx = 0;
 332         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 333         vq->vq_free_cnt = vq->vq_nentries;
 334         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 335
 336         vring_desc_init(vr->desc, size);
 337
 338         /*
 339          * Disable device(host) interrupting guest
 340          */
 341         virtqueue_disable_intr(vq);
 342 }
 343
 344 static int
 345 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 346 {
 347         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 348         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 349         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 350         unsigned int vq_size, size;
 351         struct virtio_hw *hw = dev->data->dev_private;
 352         struct virtnet_rx *rxvq = NULL;
 353         struct virtnet_tx *txvq = NULL;
 354         struct virtnet_ctl *cvq = NULL;
 355         struct virtqueue *vq;
 356         size_t sz_hdr_mz = 0;
 357         void *sw_ring = NULL;
 358         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 359         int ret;
 360
 361         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 362
 363         /*
 364          * Read the virtqueue size from the Queue Size field
 365          * Always power of 2 and if 0 virtqueue does not exist
 366          */
 367         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 368         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 369         if (vq_size == 0) {
 370                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 371                 return -EINVAL;
 372         }
 373
 374         if (!rte_is_power_of_2(vq_size)) {
 375                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 376                 return -EINVAL;
 377         }
 378
 379         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 380                  dev->data->port_id, vtpci_queue_idx);
 381
 382         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 383                                 vq_size * sizeof(struct vq_desc_extra),
 384                                 RTE_CACHE_LINE_SIZE);
 385         if (queue_type == VTNET_TQ) {
 386                 /*
 387                  * For each xmit packet, allocate a virtio_net_hdr
 388                  * and indirect ring elements
 389                  */
 390                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 391         } else if (queue_type == VTNET_CQ) {
 392                 /* Allocate a page for control vq command, data and status */
 393                 sz_hdr_mz = PAGE_SIZE;
 394         }
 395
 396         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 397                                 SOCKET_ID_ANY);
 398         if (vq == NULL) {
 399                 PMD_INIT_LOG(ERR, "can not allocate vq");
 400                 return -ENOMEM;
 401         }
 402         hw->vqs[vtpci_queue_idx] = vq;
 403
 404         vq->hw = hw;
 405         vq->vq_queue_index = vtpci_queue_idx;
 406         vq->vq_nentries = vq_size;
 407
 408         /*
 409          * Reserve a memzone for vring elements
 410          */
 411         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 412         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 413         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 414                      size, vq->vq_ring_size);
 415
 416         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 417                                          SOCKET_ID_ANY,
 418                                          0, VIRTIO_PCI_VRING_ALIGN);
 419         if (mz == NULL) {
 420                 if (rte_errno == EEXIST)
 421                         mz = rte_memzone_lookup(vq_name);
 422                 if (mz == NULL) {
 423                         ret = -ENOMEM;
 424                         goto fail_q_alloc;
 425                 }
 426         }
 427
 428         memset(mz->addr, 0, mz->len);
 429
 430         vq->vq_ring_mem = mz->phys_addr;
 431         vq->vq_ring_virt_mem = mz->addr;
 432         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 433                      (uint64_t)mz->phys_addr);
 434         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 435                      (uint64_t)(uintptr_t)mz->addr);
 436
 437         virtio_init_vring(vq);
 438
 439         if (sz_hdr_mz) {
 440                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 441                          dev->data->port_id, vtpci_queue_idx);
 442                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 443                                                      SOCKET_ID_ANY, 0,
 444                                                      RTE_CACHE_LINE_SIZE);
 445                 if (hdr_mz == NULL) {
 446                         if (rte_errno == EEXIST)
 447                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 448                         if (hdr_mz == NULL) {
 449                                 ret = -ENOMEM;
 450                                 goto fail_q_alloc;
 451                         }
 452                 }
 453         }
 454
 455         if (queue_type == VTNET_RQ) {
 456                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 457                                sizeof(vq->sw_ring[0]);
 458
 459                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 460                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
 461                 if (!sw_ring) {
 462                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 463                         ret = -ENOMEM;
 464                         goto fail_q_alloc;
 465                 }
 466
 467                 vq->sw_ring = sw_ring;
 468                 rxvq = &vq->rxq;
 469                 rxvq->vq = vq;
 470                 rxvq->port_id = dev->data->port_id;
 471                 rxvq->mz = mz;
 472         } else if (queue_type == VTNET_TQ) {
 473                 txvq = &vq->txq;
 474                 txvq->vq = vq;
 475                 txvq->port_id = dev->data->port_id;
 476                 txvq->mz = mz;
 477                 txvq->virtio_net_hdr_mz = hdr_mz;
 478                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 479         } else if (queue_type == VTNET_CQ) {
 480                 cvq = &vq->cq;
 481                 cvq->vq = vq;
 482                 cvq->mz = mz;
 483                 cvq->virtio_net_hdr_mz = hdr_mz;
 484                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 485                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 486
 487                 hw->cvq = cvq;
 488         }
 489
 490         /* For virtio_user case (that is when hw->dev is NULL), we use
 491          * virtual address. And we need properly set _offset_, please see
 492          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 493          */
 494         if (!hw->virtio_user_dev)
 495                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
 496         else {
 497                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 498                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 499                 if (queue_type == VTNET_TQ)
 500                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 501                 else if (queue_type == VTNET_CQ)
 502                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 503         }
 504
 505         if (queue_type == VTNET_TQ) {
 506                 struct virtio_tx_region *txr;
 507                 unsigned int i;
 508
 509                 txr = hdr_mz->addr;
 510                 memset(txr, 0, vq_size * sizeof(*txr));
 511                 for (i = 0; i < vq_size; i++) {
 512                         struct vring_desc *start_dp = txr[i].tx_indir;
 513
 514                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 515
 516                         /* first indirect descriptor is always the tx header */
 517                         start_dp->addr = txvq->virtio_net_hdr_mem
 518                                 + i * sizeof(*txr)
 519                                 + offsetof(struct virtio_tx_region, tx_hdr);
 520
 521                         start_dp->len = hw->vtnet_hdr_size;
 522                         start_dp->flags = VRING_DESC_F_NEXT;
 523                 }
 524         }
 525
 526         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 527                 PMD_INIT_LOG(ERR, "setup_queue failed");
 528                 return -EINVAL;
 529         }
 530
 531         return 0;
 532
 533 fail_q_alloc:
 534         rte_free(sw_ring);
 535         rte_memzone_free(hdr_mz);
 536         rte_memzone_free(mz);
 537         rte_free(vq);
 538
 539         return ret;
 540 }
 541
 542 static void
 543 virtio_free_queues(struct virtio_hw *hw)
 544 {
 545         uint16_t nr_vq = virtio_get_nr_vq(hw);
 546         struct virtqueue *vq;
 547         int queue_type;
 548         uint16_t i;
 549
 550         if (hw->vqs == NULL)
 551                 return;
 552
 553         for (i = 0; i < nr_vq; i++) {
 554                 vq = hw->vqs[i];
 555                 if (!vq)
 556                         continue;
 557
 558                 queue_type = virtio_get_queue_type(hw, i);
 559                 if (queue_type == VTNET_RQ) {
 560                         rte_free(vq->sw_ring);
 561                         rte_memzone_free(vq->rxq.mz);
 562                 } else if (queue_type == VTNET_TQ) {
 563                         rte_memzone_free(vq->txq.mz);
 564                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 565                 } else {
 566                         rte_memzone_free(vq->cq.mz);
 567                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 568                 }
 569
 570                 rte_free(vq);
 571                 hw->vqs[i] = NULL;
 572         }
 573
 574         rte_free(hw->vqs);
 575         hw->vqs = NULL;
 576 }
 577
 578 static int
 579 virtio_alloc_queues(struct rte_eth_dev *dev)
 580 {
 581         struct virtio_hw *hw = dev->data->dev_private;
 582         uint16_t nr_vq = virtio_get_nr_vq(hw);
 583         uint16_t i;
 584         int ret;
 585
 586         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 587         if (!hw->vqs) {
 588                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 589                 return -ENOMEM;
 590         }
 591
 592         for (i = 0; i < nr_vq; i++) {
 593                 ret = virtio_init_queue(dev, i);
 594                 if (ret < 0) {
 595                         virtio_free_queues(hw);
 596                         return ret;
 597                 }
 598         }
 599
 600         return 0;
 601 }
 602
 603 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 604
 605 static void
 606 virtio_dev_close(struct rte_eth_dev *dev)
 607 {
 608         struct virtio_hw *hw = dev->data->dev_private;
 609         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 610
 611         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 612
 613         /* reset the NIC */
 614         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 615                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 616         if (intr_conf->rxq)
 617                 virtio_queues_unbind_intr(dev);
 618
 619         if (intr_conf->lsc || intr_conf->rxq) {
 620                 rte_intr_disable(dev->intr_handle);
 621                 rte_intr_efd_disable(dev->intr_handle);
 622                 rte_free(dev->intr_handle->intr_vec);
 623                 dev->intr_handle->intr_vec = NULL;
 624         }
 625
 626         vtpci_reset(hw);
 627         virtio_dev_free_mbufs(dev);
 628         virtio_free_queues(hw);
 629 }
 630
 631 static void
 632 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 633 {
 634         struct virtio_hw *hw = dev->data->dev_private;
 635         struct virtio_pmd_ctrl ctrl;
 636         int dlen[1];
 637         int ret;
 638
 639         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 640                 PMD_INIT_LOG(INFO, "host does not support rx control");
 641                 return;
 642         }
 643
 644         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 645         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 646         ctrl.data[0] = 1;
 647         dlen[0] = 1;
 648
 649         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 650         if (ret)
 651                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 652 }
 653
 654 static void
 655 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 656 {
 657         struct virtio_hw *hw = dev->data->dev_private;
 658         struct virtio_pmd_ctrl ctrl;
 659         int dlen[1];
 660         int ret;
 661
 662         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 663                 PMD_INIT_LOG(INFO, "host does not support rx control");
 664                 return;
 665         }
 666
 667         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 668         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 669         ctrl.data[0] = 0;
 670         dlen[0] = 1;
 671
 672         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 673         if (ret)
 674                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 675 }
 676
 677 static void
 678 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 679 {
 680         struct virtio_hw *hw = dev->data->dev_private;
 681         struct virtio_pmd_ctrl ctrl;
 682         int dlen[1];
 683         int ret;
 684
 685         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 686                 PMD_INIT_LOG(INFO, "host does not support rx control");
 687                 return;
 688         }
 689
 690         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 691         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 692         ctrl.data[0] = 1;
 693         dlen[0] = 1;
 694
 695         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 696         if (ret)
 697                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 698 }
 699
 700 static void
 701 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 702 {
 703         struct virtio_hw *hw = dev->data->dev_private;
 704         struct virtio_pmd_ctrl ctrl;
 705         int dlen[1];
 706         int ret;
 707
 708         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 709                 PMD_INIT_LOG(INFO, "host does not support rx control");
 710                 return;
 711         }
 712
 713         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 714         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 715         ctrl.data[0] = 0;
 716         dlen[0] = 1;
 717
 718         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 719         if (ret)
 720                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 721 }
 722
 723 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 724 static int
 725 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 726 {
 727         struct virtio_hw *hw = dev->data->dev_private;
 728         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
 729                                  hw->vtnet_hdr_size;
 730         uint32_t frame_size = mtu + ether_hdr_len;
 731         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 732
 733         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 734
 735         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
 736                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 737                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 738                 return -EINVAL;
 739         }
 740         return 0;
 741 }
 742
 743 static int
 744 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 745 {
 746         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 747         struct virtqueue *vq = rxvq->vq;
 748
 749         virtqueue_enable_intr(vq);
 750         return 0;
 751 }
 752
 753 static int
 754 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 755 {
 756         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 757         struct virtqueue *vq = rxvq->vq;
 758
 759         virtqueue_disable_intr(vq);
 760         return 0;
 761 }
 762
 763 /*
 764  * dev_ops for virtio, bare necessities for basic operation
 765  */
 766 static const struct eth_dev_ops virtio_eth_dev_ops = {
 767         .dev_configure           = virtio_dev_configure,
 768         .dev_start               = virtio_dev_start,
 769         .dev_stop                = virtio_dev_stop,
 770         .dev_close               = virtio_dev_close,
 771         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 772         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 773         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 774         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 775         .mtu_set                 = virtio_mtu_set,
 776         .dev_infos_get           = virtio_dev_info_get,
 777         .stats_get               = virtio_dev_stats_get,
 778         .xstats_get              = virtio_dev_xstats_get,
 779         .xstats_get_names        = virtio_dev_xstats_get_names,
 780         .stats_reset             = virtio_dev_stats_reset,
 781         .xstats_reset            = virtio_dev_stats_reset,
 782         .link_update             = virtio_dev_link_update,
 783         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 784         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 785         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 786         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 787         .rx_queue_release        = virtio_dev_queue_release,
 788         .rx_descriptor_done      = virtio_dev_rx_queue_done,
 789         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 790         .tx_queue_release        = virtio_dev_queue_release,
 791         /* collect stats per queue */
 792         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 793         .vlan_filter_set         = virtio_vlan_filter_set,
 794         .mac_addr_add            = virtio_mac_addr_add,
 795         .mac_addr_remove         = virtio_mac_addr_remove,
 796         .mac_addr_set            = virtio_mac_addr_set,
 797 };
 798
 799 static inline int
 800 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 801                                 struct rte_eth_link *link)
 802 {
 803         struct rte_eth_link *dst = link;
 804         struct rte_eth_link *src = &(dev->data->dev_link);
 805
 806         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 807                         *(uint64_t *)src) == 0)
 808                 return -1;
 809
 810         return 0;
 811 }
 812
 813 /**
 814  * Atomically writes the link status information into global
 815  * structure rte_eth_dev.
 816  *
 817  * @param dev
 818  *   - Pointer to the structure rte_eth_dev to read from.
 819  *   - Pointer to the buffer to be saved with the link status.
 820  *
 821  * @return
 822  *   - On success, zero.
 823  *   - On failure, negative value.
 824  */
 825 static inline int
 826 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 827                 struct rte_eth_link *link)
 828 {
 829         struct rte_eth_link *dst = &(dev->data->dev_link);
 830         struct rte_eth_link *src = link;
 831
 832         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 833                                         *(uint64_t *)src) == 0)
 834                 return -1;
 835
 836         return 0;
 837 }
 838
 839 static void
 840 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 841 {
 842         unsigned i;
 843
 844         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 845                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 846                 if (txvq == NULL)
 847                         continue;
 848
 849                 stats->opackets += txvq->stats.packets;
 850                 stats->obytes += txvq->stats.bytes;
 851                 stats->oerrors += txvq->stats.errors;
 852
 853                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 854                         stats->q_opackets[i] = txvq->stats.packets;
 855                         stats->q_obytes[i] = txvq->stats.bytes;
 856                 }
 857         }
 858
 859         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 860                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 861                 if (rxvq == NULL)
 862                         continue;
 863
 864                 stats->ipackets += rxvq->stats.packets;
 865                 stats->ibytes += rxvq->stats.bytes;
 866                 stats->ierrors += rxvq->stats.errors;
 867
 868                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 869                         stats->q_ipackets[i] = rxvq->stats.packets;
 870                         stats->q_ibytes[i] = rxvq->stats.bytes;
 871                 }
 872         }
 873
 874         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 875 }
 876
 877 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 878                                        struct rte_eth_xstat_name *xstats_names,
 879                                        __rte_unused unsigned limit)
 880 {
 881         unsigned i;
 882         unsigned count = 0;
 883         unsigned t;
 884
 885         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 886                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 887
 888         if (xstats_names != NULL) {
 889                 /* Note: limit checked in rte_eth_xstats_names() */
 890
 891                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 892                         struct virtqueue *rxvq = dev->data->rx_queues[i];
 893                         if (rxvq == NULL)
 894                                 continue;
 895                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 896                                 snprintf(xstats_names[count].name,
 897                                         sizeof(xstats_names[count].name),
 898                                         "rx_q%u_%s", i,
 899                                         rte_virtio_rxq_stat_strings[t].name);
 900                                 count++;
 901                         }
 902                 }
 903
 904                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 905                         struct virtqueue *txvq = dev->data->tx_queues[i];
 906                         if (txvq == NULL)
 907                                 continue;
 908                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 909                                 snprintf(xstats_names[count].name,
 910                                         sizeof(xstats_names[count].name),
 911                                         "tx_q%u_%s", i,
 912                                         rte_virtio_txq_stat_strings[t].name);
 913                                 count++;
 914                         }
 915                 }
 916                 return count;
 917         }
 918         return nstats;
 919 }
 920
 921 static int
 922 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 923                       unsigned n)
 924 {
 925         unsigned i;
 926         unsigned count = 0;
 927
 928         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 929                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 930
 931         if (n < nstats)
 932                 return nstats;
 933
 934         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 935                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 936
 937                 if (rxvq == NULL)
 938                         continue;
 939
 940                 unsigned t;
 941
 942                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 943                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 944                                 rte_virtio_rxq_stat_strings[t].offset);
 945                         xstats[count].id = count;
 946                         count++;
 947                 }
 948         }
 949
 950         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 951                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 952
 953                 if (txvq == NULL)
 954                         continue;
 955
 956                 unsigned t;
 957
 958                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 959                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 960                                 rte_virtio_txq_stat_strings[t].offset);
 961                         xstats[count].id = count;
 962                         count++;
 963                 }
 964         }
 965
 966         return count;
 967 }
 968
 969 static int
 970 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 971 {
 972         virtio_update_stats(dev, stats);
 973
 974         return 0;
 975 }
 976
 977 static void
 978 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 979 {
 980         unsigned int i;
 981
 982         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 983                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 984                 if (txvq == NULL)
 985                         continue;
 986
 987                 txvq->stats.packets = 0;
 988                 txvq->stats.bytes = 0;
 989                 txvq->stats.errors = 0;
 990                 txvq->stats.multicast = 0;
 991                 txvq->stats.broadcast = 0;
 992                 memset(txvq->stats.size_bins, 0,
 993                        sizeof(txvq->stats.size_bins[0]) * 8);
 994         }
 995
 996         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 997                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 998                 if (rxvq == NULL)
 999                         continue;
1000
1001                 rxvq->stats.packets = 0;
1002                 rxvq->stats.bytes = 0;
1003                 rxvq->stats.errors = 0;
1004                 rxvq->stats.multicast = 0;
1005                 rxvq->stats.broadcast = 0;
1006                 memset(rxvq->stats.size_bins, 0,
1007                        sizeof(rxvq->stats.size_bins[0]) * 8);
1008         }
1009 }
1010
1011 static void
1012 virtio_set_hwaddr(struct virtio_hw *hw)
1013 {
1014         vtpci_write_dev_config(hw,
1015                         offsetof(struct virtio_net_config, mac),
1016                         &hw->mac_addr, ETHER_ADDR_LEN);
1017 }
1018
1019 static void
1020 virtio_get_hwaddr(struct virtio_hw *hw)
1021 {
1022         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1023                 vtpci_read_dev_config(hw,
1024                         offsetof(struct virtio_net_config, mac),
1025                         &hw->mac_addr, ETHER_ADDR_LEN);
1026         } else {
1027                 eth_random_addr(&hw->mac_addr[0]);
1028                 virtio_set_hwaddr(hw);
1029         }
1030 }
1031
1032 static int
1033 virtio_mac_table_set(struct virtio_hw *hw,
1034                      const struct virtio_net_ctrl_mac *uc,
1035                      const struct virtio_net_ctrl_mac *mc)
1036 {
1037         struct virtio_pmd_ctrl ctrl;
1038         int err, len[2];
1039
1040         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1041                 PMD_DRV_LOG(INFO, "host does not support mac table");
1042                 return -1;
1043         }
1044
1045         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1046         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1047
1048         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1049         memcpy(ctrl.data, uc, len[0]);
1050
1051         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1052         memcpy(ctrl.data + len[0], mc, len[1]);
1053
1054         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1055         if (err != 0)
1056                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1057         return err;
1058 }
1059
1060 static int
1061 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1062                     uint32_t index, uint32_t vmdq __rte_unused)
1063 {
1064         struct virtio_hw *hw = dev->data->dev_private;
1065         const struct ether_addr *addrs = dev->data->mac_addrs;
1066         unsigned int i;
1067         struct virtio_net_ctrl_mac *uc, *mc;
1068
1069         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1070                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1071                 return -EINVAL;
1072         }
1073
1074         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1075         uc->entries = 0;
1076         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1077         mc->entries = 0;
1078
1079         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1080                 const struct ether_addr *addr
1081                         = (i == index) ? mac_addr : addrs + i;
1082                 struct virtio_net_ctrl_mac *tbl
1083                         = is_multicast_ether_addr(addr) ? mc : uc;
1084
1085                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1086         }
1087
1088         return virtio_mac_table_set(hw, uc, mc);
1089 }
1090
1091 static void
1092 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1093 {
1094         struct virtio_hw *hw = dev->data->dev_private;
1095         struct ether_addr *addrs = dev->data->mac_addrs;
1096         struct virtio_net_ctrl_mac *uc, *mc;
1097         unsigned int i;
1098
1099         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1100                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1101                 return;
1102         }
1103
1104         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1105         uc->entries = 0;
1106         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1107         mc->entries = 0;
1108
1109         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1110                 struct virtio_net_ctrl_mac *tbl;
1111
1112                 if (i == index || is_zero_ether_addr(addrs + i))
1113                         continue;
1114
1115                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1116                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1117         }
1118
1119         virtio_mac_table_set(hw, uc, mc);
1120 }
1121
1122 static void
1123 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1124 {
1125         struct virtio_hw *hw = dev->data->dev_private;
1126
1127         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1128
1129         /* Use atomic update if available */
1130         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1131                 struct virtio_pmd_ctrl ctrl;
1132                 int len = ETHER_ADDR_LEN;
1133
1134                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1135                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1136
1137                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1138                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1139         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1140                 virtio_set_hwaddr(hw);
1141 }
1142
1143 static int
1144 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1145 {
1146         struct virtio_hw *hw = dev->data->dev_private;
1147         struct virtio_pmd_ctrl ctrl;
1148         int len;
1149
1150         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1151                 return -ENOTSUP;
1152
1153         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1154         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1155         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1156         len = sizeof(vlan_id);
1157
1158         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1159 }
1160
1161 static int
1162 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1163 {
1164         uint64_t host_features;
1165
1166         /* Prepare guest_features: feature that driver wants to support */
1167         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1168                 req_features);
1169
1170         /* Read device(host) feature bits */
1171         host_features = VTPCI_OPS(hw)->get_features(hw);
1172         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1173                 host_features);
1174
1175         /* If supported, ensure MTU value is valid before acknowledging it. */
1176         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1177                 struct virtio_net_config config;
1178
1179                 vtpci_read_dev_config(hw,
1180                         offsetof(struct virtio_net_config, mtu),
1181                         &config.mtu, sizeof(config.mtu));
1182
1183                 if (config.mtu < ETHER_MIN_MTU)
1184                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1185         }
1186
1187         /*
1188          * Negotiate features: Subset of device feature bits are written back
1189          * guest feature bits.
1190          */
1191         hw->guest_features = req_features;
1192         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1193         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1194                 hw->guest_features);
1195
1196         if (hw->modern) {
1197                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1198                         PMD_INIT_LOG(ERR,
1199                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1200                         return -1;
1201                 }
1202                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1203                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1204                         PMD_INIT_LOG(ERR,
1205                                 "failed to set FEATURES_OK status!");
1206                         return -1;
1207                 }
1208         }
1209
1210         hw->req_guest_features = req_features;
1211
1212         return 0;
1213 }
1214
1215 /*
1216  * Process Virtio Config changed interrupt and call the callback
1217  * if link state changed.
1218  */
1219 void
1220 virtio_interrupt_handler(void *param)
1221 {
1222         struct rte_eth_dev *dev = param;
1223         struct virtio_hw *hw = dev->data->dev_private;
1224         uint8_t isr;
1225
1226         /* Read interrupt status which clears interrupt */
1227         isr = vtpci_isr(hw);
1228         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1229
1230         if (rte_intr_enable(dev->intr_handle) < 0)
1231                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1232
1233         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1234                 if (virtio_dev_link_update(dev, 0) == 0)
1235                         _rte_eth_dev_callback_process(dev,
1236                                                       RTE_ETH_EVENT_INTR_LSC,
1237                                                       NULL, NULL);
1238         }
1239
1240 }
1241
1242 /* set rx and tx handlers according to what is supported */
1243 static void
1244 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1245 {
1246         struct virtio_hw *hw = eth_dev->data->dev_private;
1247
1248         if (hw->use_simple_rx) {
1249                 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1250                         eth_dev->data->port_id);
1251                 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1252         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1253                 PMD_INIT_LOG(INFO,
1254                         "virtio: using mergeable buffer Rx path on port %u",
1255                         eth_dev->data->port_id);
1256                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1257         } else {
1258                 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1259                         eth_dev->data->port_id);
1260                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1261         }
1262
1263         if (hw->use_simple_tx) {
1264                 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1265                         eth_dev->data->port_id);
1266                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1267         } else {
1268                 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1269                         eth_dev->data->port_id);
1270                 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1271         }
1272 }
1273
1274 /* Only support 1:1 queue/interrupt mapping so far.
1275  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1276  * interrupt vectors (<N+1).
1277  */
1278 static int
1279 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1280 {
1281         uint32_t i;
1282         struct virtio_hw *hw = dev->data->dev_private;
1283
1284         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1285         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1286                 dev->intr_handle->intr_vec[i] = i + 1;
1287                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1288                                                  VIRTIO_MSI_NO_VECTOR) {
1289                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1290                         return -EBUSY;
1291                 }
1292         }
1293
1294         return 0;
1295 }
1296
1297 static void
1298 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1299 {
1300         uint32_t i;
1301         struct virtio_hw *hw = dev->data->dev_private;
1302
1303         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1304         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1305                 VTPCI_OPS(hw)->set_queue_irq(hw,
1306                                              hw->vqs[i * VTNET_CQ],
1307                                              VIRTIO_MSI_NO_VECTOR);
1308 }
1309
1310 static int
1311 virtio_configure_intr(struct rte_eth_dev *dev)
1312 {
1313         struct virtio_hw *hw = dev->data->dev_private;
1314
1315         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1316                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1317                 return -ENOTSUP;
1318         }
1319
1320         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1321                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1322                 return -1;
1323         }
1324
1325         if (!dev->intr_handle->intr_vec) {
1326                 dev->intr_handle->intr_vec =
1327                         rte_zmalloc("intr_vec",
1328                                     hw->max_queue_pairs * sizeof(int), 0);
1329                 if (!dev->intr_handle->intr_vec) {
1330                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1331                                      hw->max_queue_pairs);
1332                         return -ENOMEM;
1333                 }
1334         }
1335
1336         /* Re-register callback to update max_intr */
1337         rte_intr_callback_unregister(dev->intr_handle,
1338                                      virtio_interrupt_handler,
1339                                      dev);
1340         rte_intr_callback_register(dev->intr_handle,
1341                                    virtio_interrupt_handler,
1342                                    dev);
1343
1344         /* DO NOT try to remove this! This function will enable msix, or QEMU
1345          * will encounter SIGSEGV when DRIVER_OK is sent.
1346          * And for legacy devices, this should be done before queue/vec binding
1347          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1348          * (22) will be ignored.
1349          */
1350         if (rte_intr_enable(dev->intr_handle) < 0) {
1351                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1352                 return -1;
1353         }
1354
1355         if (virtio_queues_bind_intr(dev) < 0) {
1356                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1357                 return -1;
1358         }
1359
1360         return 0;
1361 }
1362
1363 /* reset device and renegotiate features if needed */
1364 static int
1365 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1366 {
1367         struct virtio_hw *hw = eth_dev->data->dev_private;
1368         struct virtio_net_config *config;
1369         struct virtio_net_config local_config;
1370         struct rte_pci_device *pci_dev = NULL;
1371         int ret;
1372
1373         /* Reset the device although not necessary at startup */
1374         vtpci_reset(hw);
1375
1376         /* Tell the host we've noticed this device. */
1377         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1378
1379         /* Tell the host we've known how to drive the device. */
1380         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1381         if (virtio_negotiate_features(hw, req_features) < 0)
1382                 return -1;
1383
1384         if (!hw->virtio_user_dev) {
1385                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1386                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1387         }
1388
1389         /* If host does not support both status and MSI-X then disable LSC */
1390         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) && hw->use_msix)
1391                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1392         else
1393                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1394
1395         /* Setting up rx_header size for the device */
1396         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1397             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1398                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1399         else
1400                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1401
1402         /* Copy the permanent MAC address to: virtio_hw */
1403         virtio_get_hwaddr(hw);
1404         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1405                         &eth_dev->data->mac_addrs[0]);
1406         PMD_INIT_LOG(DEBUG,
1407                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1408                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1409                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1410
1411         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1412                 config = &local_config;
1413
1414                 vtpci_read_dev_config(hw,
1415                         offsetof(struct virtio_net_config, mac),
1416                         &config->mac, sizeof(config->mac));
1417
1418                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1419                         vtpci_read_dev_config(hw,
1420                                 offsetof(struct virtio_net_config, status),
1421                                 &config->status, sizeof(config->status));
1422                 } else {
1423                         PMD_INIT_LOG(DEBUG,
1424                                      "VIRTIO_NET_F_STATUS is not supported");
1425                         config->status = 0;
1426                 }
1427
1428                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1429                         vtpci_read_dev_config(hw,
1430                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1431                                 &config->max_virtqueue_pairs,
1432                                 sizeof(config->max_virtqueue_pairs));
1433                 } else {
1434                         PMD_INIT_LOG(DEBUG,
1435                                      "VIRTIO_NET_F_MQ is not supported");
1436                         config->max_virtqueue_pairs = 1;
1437                 }
1438
1439                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1440
1441                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1442                         vtpci_read_dev_config(hw,
1443                                 offsetof(struct virtio_net_config, mtu),
1444                                 &config->mtu,
1445                                 sizeof(config->mtu));
1446
1447                         /*
1448                          * MTU value has already been checked at negotiation
1449                          * time, but check again in case it has changed since
1450                          * then, which should not happen.
1451                          */
1452                         if (config->mtu < ETHER_MIN_MTU) {
1453                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1454                                                 config->mtu);
1455                                 return -1;
1456                         }
1457
1458                         hw->max_mtu = config->mtu;
1459                         /* Set initial MTU to maximum one supported by vhost */
1460                         eth_dev->data->mtu = config->mtu;
1461
1462                 } else {
1463                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1464                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1465                 }
1466
1467                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1468                                 config->max_virtqueue_pairs);
1469                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1470                 PMD_INIT_LOG(DEBUG,
1471                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1472                                 config->mac[0], config->mac[1],
1473                                 config->mac[2], config->mac[3],
1474                                 config->mac[4], config->mac[5]);
1475         } else {
1476                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1477                 hw->max_queue_pairs = 1;
1478         }
1479
1480         ret = virtio_alloc_queues(eth_dev);
1481         if (ret < 0)
1482                 return ret;
1483
1484         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1485                 if (virtio_configure_intr(eth_dev) < 0) {
1486                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1487                         return -1;
1488                 }
1489         }
1490
1491         vtpci_reinit_complete(hw);
1492
1493         if (pci_dev)
1494                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1495                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1496                         pci_dev->id.device_id);
1497
1498         return 0;
1499 }
1500
1501 /*
1502  * Remap the PCI device again (IO port map for legacy device and
1503  * memory map for modern device), so that the secondary process
1504  * could have the PCI initiated correctly.
1505  */
1506 static int
1507 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1508 {
1509         if (hw->modern) {
1510                 /*
1511                  * We don't have to re-parse the PCI config space, since
1512                  * rte_pci_map_device() makes sure the mapped address
1513                  * in secondary process would equal to the one mapped in
1514                  * the primary process: error will be returned if that
1515                  * requirement is not met.
1516                  *
1517                  * That said, we could simply reuse all cap pointers
1518                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1519                  * primary process, which is stored in shared memory.
1520                  */
1521                 if (rte_pci_map_device(pci_dev)) {
1522                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1523                         return -1;
1524                 }
1525         } else {
1526                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1527                         return -1;
1528         }
1529
1530         return 0;
1531 }
1532
1533 static void
1534 virtio_set_vtpci_ops(struct virtio_hw *hw)
1535 {
1536 #ifdef RTE_VIRTIO_USER
1537         if (hw->virtio_user_dev)
1538                 VTPCI_OPS(hw) = &virtio_user_ops;
1539         else
1540 #endif
1541         if (hw->modern)
1542                 VTPCI_OPS(hw) = &modern_ops;
1543         else
1544                 VTPCI_OPS(hw) = &legacy_ops;
1545 }
1546
1547 /*
1548  * This function is based on probe() function in virtio_pci.c
1549  * It returns 0 on success.
1550  */
1551 int
1552 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1553 {
1554         struct virtio_hw *hw = eth_dev->data->dev_private;
1555         int ret;
1556
1557         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1558
1559         eth_dev->dev_ops = &virtio_eth_dev_ops;
1560
1561         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1562                 if (!hw->virtio_user_dev) {
1563                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1564                         if (ret)
1565                                 return ret;
1566                 }
1567
1568                 virtio_set_vtpci_ops(hw);
1569                 set_rxtx_funcs(eth_dev);
1570
1571                 return 0;
1572         }
1573
1574         /* Allocate memory for storing MAC addresses */
1575         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1576         if (eth_dev->data->mac_addrs == NULL) {
1577                 PMD_INIT_LOG(ERR,
1578                         "Failed to allocate %d bytes needed to store MAC addresses",
1579                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1580                 return -ENOMEM;
1581         }
1582
1583         hw->port_id = eth_dev->data->port_id;
1584         /* For virtio_user case the hw->virtio_user_dev is populated by
1585          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1586          */
1587         if (!hw->virtio_user_dev) {
1588                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1589                 if (ret)
1590                         return ret;
1591         }
1592
1593         /* reset device and negotiate default features */
1594         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1595         if (ret < 0)
1596                 return ret;
1597
1598         /* Setup interrupt callback  */
1599         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1600                 rte_intr_callback_register(eth_dev->intr_handle,
1601                         virtio_interrupt_handler, eth_dev);
1602
1603         return 0;
1604 }
1605
1606 static int
1607 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1608 {
1609         PMD_INIT_FUNC_TRACE();
1610
1611         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1612                 return -EPERM;
1613
1614         virtio_dev_stop(eth_dev);
1615         virtio_dev_close(eth_dev);
1616
1617         eth_dev->dev_ops = NULL;
1618         eth_dev->tx_pkt_burst = NULL;
1619         eth_dev->rx_pkt_burst = NULL;
1620
1621         rte_free(eth_dev->data->mac_addrs);
1622         eth_dev->data->mac_addrs = NULL;
1623
1624         /* reset interrupt callback  */
1625         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1626                 rte_intr_callback_unregister(eth_dev->intr_handle,
1627                                                 virtio_interrupt_handler,
1628                                                 eth_dev);
1629         if (eth_dev->device)
1630                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1631
1632         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1633
1634         return 0;
1635 }
1636
1637 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1638         struct rte_pci_device *pci_dev)
1639 {
1640         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1641                 eth_virtio_dev_init);
1642 }
1643
1644 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1645 {
1646         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1647 }
1648
1649 static struct rte_pci_driver rte_virtio_pmd = {
1650         .driver = {
1651                 .name = "net_virtio",
1652         },
1653         .id_table = pci_id_virtio_map,
1654         .drv_flags = 0,
1655         .probe = eth_virtio_pci_probe,
1656         .remove = eth_virtio_pci_remove,
1657 };
1658
1659 RTE_INIT(rte_virtio_pmd_init);
1660 static void
1661 rte_virtio_pmd_init(void)
1662 {
1663         if (rte_eal_iopl_init() != 0) {
1664                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1665                 return;
1666         }
1667
1668         rte_pci_register(&rte_virtio_pmd);
1669 }
1670
1671 /*
1672  * Configure virtio device
1673  * It returns 0 on success.
1674  */
1675 static int
1676 virtio_dev_configure(struct rte_eth_dev *dev)
1677 {
1678         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1679         struct virtio_hw *hw = dev->data->dev_private;
1680         uint64_t req_features;
1681         int ret;
1682
1683         PMD_INIT_LOG(DEBUG, "configure");
1684         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1685
1686         if (dev->data->dev_conf.intr_conf.rxq) {
1687                 ret = virtio_init_device(dev, hw->req_guest_features);
1688                 if (ret < 0)
1689                         return ret;
1690         }
1691
1692         /* The name hw_ip_checksum is a bit confusing since it can be
1693          * set by the application to request L3 and/or L4 checksums. In
1694          * case of virtio, only L4 checksum is supported.
1695          */
1696         if (rxmode->hw_ip_checksum)
1697                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1698
1699         if (rxmode->enable_lro)
1700                 req_features |=
1701                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1702                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1703
1704         /* if request features changed, reinit the device */
1705         if (req_features != hw->req_guest_features) {
1706                 ret = virtio_init_device(dev, req_features);
1707                 if (ret < 0)
1708                         return ret;
1709         }
1710
1711         if (rxmode->hw_ip_checksum &&
1712                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1713                 PMD_DRV_LOG(ERR,
1714                         "rx checksum not available on this host");
1715                 return -ENOTSUP;
1716         }
1717
1718         if (rxmode->enable_lro &&
1719                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1720                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1721                 PMD_DRV_LOG(ERR,
1722                         "Large Receive Offload not available on this host");
1723                 return -ENOTSUP;
1724         }
1725
1726         /* start control queue */
1727         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1728                 virtio_dev_cq_start(dev);
1729
1730         hw->vlan_strip = rxmode->hw_vlan_strip;
1731
1732         if (rxmode->hw_vlan_filter
1733             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1734                 PMD_DRV_LOG(ERR,
1735                             "vlan filtering not available on this host");
1736                 return -ENOTSUP;
1737         }
1738
1739         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1740                 /* Enable vector (0) for Link State Intrerrupt */
1741                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1742                                 VIRTIO_MSI_NO_VECTOR) {
1743                         PMD_DRV_LOG(ERR, "failed to set config vector");
1744                         return -EBUSY;
1745                 }
1746
1747         hw->use_simple_rx = 1;
1748         hw->use_simple_tx = 1;
1749
1750 #if defined RTE_ARCH_ARM64 || defined CONFIG_RTE_ARCH_ARM
1751         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1752                 hw->use_simple_rx = 0;
1753                 hw->use_simple_tx = 0;
1754         }
1755 #endif
1756         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1757                 hw->use_simple_rx = 0;
1758                 hw->use_simple_tx = 0;
1759         }
1760
1761         if (rxmode->hw_ip_checksum)
1762                 hw->use_simple_rx = 0;
1763
1764         return 0;
1765 }
1766
1767
1768 static int
1769 virtio_dev_start(struct rte_eth_dev *dev)
1770 {
1771         uint16_t nb_queues, i;
1772         struct virtnet_rx *rxvq;
1773         struct virtnet_tx *txvq __rte_unused;
1774         struct virtio_hw *hw = dev->data->dev_private;
1775         int ret;
1776
1777         /* Finish the initialization of the queues */
1778         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1779                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1780                 if (ret < 0)
1781                         return ret;
1782         }
1783         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1784                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1785                 if (ret < 0)
1786                         return ret;
1787         }
1788
1789         /* check if lsc interrupt feature is enabled */
1790         if (dev->data->dev_conf.intr_conf.lsc) {
1791                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1792                         PMD_DRV_LOG(ERR, "link status not supported by host");
1793                         return -ENOTSUP;
1794                 }
1795         }
1796
1797         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1798          * in device configure, but it could be unmapped  when device is
1799          * stopped.
1800          */
1801         if (dev->data->dev_conf.intr_conf.lsc ||
1802             dev->data->dev_conf.intr_conf.rxq) {
1803                 rte_intr_disable(dev->intr_handle);
1804
1805                 if (rte_intr_enable(dev->intr_handle) < 0) {
1806                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1807                         return -EIO;
1808                 }
1809         }
1810
1811         /*Notify the backend
1812          *Otherwise the tap backend might already stop its queue due to fullness.
1813          *vhost backend will have no chance to be waked up
1814          */
1815         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1816         if (hw->max_queue_pairs > 1) {
1817                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1818                         return -EINVAL;
1819         }
1820
1821         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1822
1823         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1824                 rxvq = dev->data->rx_queues[i];
1825                 /* Flush the old packets */
1826                 virtqueue_flush(rxvq->vq);
1827                 virtqueue_notify(rxvq->vq);
1828         }
1829
1830         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1831                 txvq = dev->data->tx_queues[i];
1832                 virtqueue_notify(txvq->vq);
1833         }
1834
1835         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1836
1837         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1838                 rxvq = dev->data->rx_queues[i];
1839                 VIRTQUEUE_DUMP(rxvq->vq);
1840         }
1841
1842         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1843                 txvq = dev->data->tx_queues[i];
1844                 VIRTQUEUE_DUMP(txvq->vq);
1845         }
1846
1847         set_rxtx_funcs(dev);
1848         hw->started = 1;
1849
1850         /* Initialize Link state */
1851         virtio_dev_link_update(dev, 0);
1852
1853         return 0;
1854 }
1855
1856 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1857 {
1858         struct rte_mbuf *buf;
1859         int i, mbuf_num = 0;
1860
1861         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1862                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1863
1864                 PMD_INIT_LOG(DEBUG,
1865                              "Before freeing rxq[%d] used and unused buf", i);
1866                 VIRTQUEUE_DUMP(rxvq->vq);
1867
1868                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1869                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1870                         rte_pktmbuf_free(buf);
1871                         mbuf_num++;
1872                 }
1873
1874                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1875                 PMD_INIT_LOG(DEBUG,
1876                              "After freeing rxq[%d] used and unused buf", i);
1877                 VIRTQUEUE_DUMP(rxvq->vq);
1878         }
1879
1880         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1881                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1882
1883                 PMD_INIT_LOG(DEBUG,
1884                              "Before freeing txq[%d] used and unused bufs",
1885                              i);
1886                 VIRTQUEUE_DUMP(txvq->vq);
1887
1888                 mbuf_num = 0;
1889                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1890                         rte_pktmbuf_free(buf);
1891                         mbuf_num++;
1892                 }
1893
1894                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1895                 PMD_INIT_LOG(DEBUG,
1896                              "After freeing txq[%d] used and unused buf", i);
1897                 VIRTQUEUE_DUMP(txvq->vq);
1898         }
1899 }
1900
1901 /*
1902  * Stop device: disable interrupt and mark link down
1903  */
1904 static void
1905 virtio_dev_stop(struct rte_eth_dev *dev)
1906 {
1907         struct virtio_hw *hw = dev->data->dev_private;
1908         struct rte_eth_link link;
1909         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1910
1911         PMD_INIT_LOG(DEBUG, "stop");
1912
1913         if (intr_conf->lsc || intr_conf->rxq)
1914                 rte_intr_disable(dev->intr_handle);
1915
1916         hw->started = 0;
1917         memset(&link, 0, sizeof(link));
1918         virtio_dev_atomic_write_link_status(dev, &link);
1919 }
1920
1921 static int
1922 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1923 {
1924         struct rte_eth_link link, old;
1925         uint16_t status;
1926         struct virtio_hw *hw = dev->data->dev_private;
1927         memset(&link, 0, sizeof(link));
1928         virtio_dev_atomic_read_link_status(dev, &link);
1929         old = link;
1930         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1931         link.link_speed  = SPEED_10G;
1932
1933         if (hw->started == 0) {
1934                 link.link_status = ETH_LINK_DOWN;
1935         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1936                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1937                 vtpci_read_dev_config(hw,
1938                                 offsetof(struct virtio_net_config, status),
1939                                 &status, sizeof(status));
1940                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1941                         link.link_status = ETH_LINK_DOWN;
1942                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1943                                      dev->data->port_id);
1944                 } else {
1945                         link.link_status = ETH_LINK_UP;
1946                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1947                                      dev->data->port_id);
1948                 }
1949         } else {
1950                 link.link_status = ETH_LINK_UP;
1951         }
1952         virtio_dev_atomic_write_link_status(dev, &link);
1953
1954         return (old.link_status == link.link_status) ? -1 : 0;
1955 }
1956
1957 static int
1958 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1959 {
1960         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1961         struct virtio_hw *hw = dev->data->dev_private;
1962
1963         if (mask & ETH_VLAN_FILTER_MASK) {
1964                 if (rxmode->hw_vlan_filter &&
1965                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1966
1967                         PMD_DRV_LOG(NOTICE,
1968                                 "vlan filtering not available on this host");
1969
1970                         return -ENOTSUP;
1971                 }
1972         }
1973
1974         if (mask & ETH_VLAN_STRIP_MASK)
1975                 hw->vlan_strip = rxmode->hw_vlan_strip;
1976
1977         return 0;
1978 }
1979
1980 static void
1981 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1982 {
1983         uint64_t tso_mask, host_features;
1984         struct virtio_hw *hw = dev->data->dev_private;
1985
1986         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
1987
1988         dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
1989         dev_info->max_rx_queues =
1990                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
1991         dev_info->max_tx_queues =
1992                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
1993         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1994         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1995         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1996         dev_info->default_txconf = (struct rte_eth_txconf) {
1997                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1998         };
1999
2000         host_features = VTPCI_OPS(hw)->get_features(hw);
2001         dev_info->rx_offload_capa = 0;
2002         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2003                 dev_info->rx_offload_capa |=
2004                         DEV_RX_OFFLOAD_TCP_CKSUM |
2005                         DEV_RX_OFFLOAD_UDP_CKSUM;
2006         }
2007         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2008                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2009         if ((host_features & tso_mask) == tso_mask)
2010                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2011
2012         dev_info->tx_offload_capa = 0;
2013         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2014                 dev_info->tx_offload_capa |=
2015                         DEV_TX_OFFLOAD_UDP_CKSUM |
2016                         DEV_TX_OFFLOAD_TCP_CKSUM;
2017         }
2018         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2019                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2020         if ((hw->guest_features & tso_mask) == tso_mask)
2021                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2022 }
2023
2024 /*
2025  * It enables testpmd to collect per queue stats.
2026  */
2027 static int
2028 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2029 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2030 __rte_unused uint8_t is_rx)
2031 {
2032         return 0;
2033 }
2034
2035 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2036 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2037 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");